Elastic Net Research Articles

Predicting lymph node metastasis in colorectal cancer patients: development and validation of a column chart model

Lymph node metastasis (LNM) is one of the crucial factors in determining the optimal treatment approach for colorectal cancer. The objective of this study was to establish and validate a column chart for predicting LNM in colon cancer patients. We extracted a total of 83,430 cases of colon cancer from the Surveillance, Epidemiology, and End Results (SEER) database, spanning the years 2010–2017. These cases were divided into a training group and a testing group in a 7:3 ratio. An additional 8545 patients from the years 2018–2019 were used for external validation. Univariate and multivariate logistic regression models were employed in the training set to identify predictive factors. Models were developed using logistic regression, LASSO regression, ridge regression, and elastic net regression algorithms. Model performance was quantified by calculating the area under the ROC curve (AUC) and its corresponding 95% confidence interval. The results demonstrated that tumor location, grade, age, tumor size, T stage, race, and CEA were independent predictors of LNM in CRC patients. The logistic regression model yielded an AUC of 0.708 (0.7038–0.7122), outperforming ridge regression and achieving similar AUC values as LASSO regression and elastic net regression. Based on the logistic regression algorithm, we constructed a column chart for predicting LNM in CRC patients. Further subgroup analysis based on gender, age, and grade indicated that the logistic prediction model exhibited good adaptability across all subgroups. Our column chart displayed excellent predictive capability and serves as a useful tool for clinicians in predicting LNM in colorectal cancer patients.

Predicting subacute ruminal acidosis from milk mid-infrared estimated fatty acids and machine learning on Canadian commercial dairy herds

Our objective was to validate the possibility of detecting SARA from milk Fourier transform mid-infrared spectroscopy estimated fatty acids (FA) and machine learning. Subacute ruminal acidosis is a common condition in modern commercial dairy herds for which the diagnostic remains challenging due to its symptoms often being subtle, nonexclusive, and not immediately apparent. This observational study aimed at evaluating the possibility of predicting SARA by developing machine learning models to be applied to farm data and to provide an estimated portrait of SARA prevalence in commercial dairy herds. A first data set composed of 488 milk samples of 67 cows (initial DIM = 8.5 ± 6.18; mean ± SD) from 7 commercial dairy farms and their corresponding SARA classification (SARA+ if rumen pH <6.0 for 300 min, else SARA-) was used for the development of machine learning models. Three sets of predictive variables: i) milk major components (MMC), ii) milk FA (MFA), and iii) MMC combined with MFA (MMCFA) were submitted to 3 different algorithms, namely Elastic net (EN), Extreme gradient boosting (XGB), and Partial least squares (PLS), and evaluated using 3 different scenarios of cross-validation. Accuracy, sensitivity, and specificity of the resulting 27 models were analyzed using a linear mixed model. Model performance was not significantly affected by the choice of algorithm. Model performance was improved by including fatty acids estimations (MFA and MMCFA as opposed to MMC alone). Based on these results, one model was selected (algorithm: EN; predictive variables: MMCFA; 60.4, 65.4, and 55.3% of accuracy, sensitivity, and specificity, respectively) and applied to a large data set comprising the first test-day record (milk major components and FA within the first 70 DIM of 211,972 Holstein cows (219,503 samples) collected from 3001 commercial dairy herds. Based on this analysis, the within-herd SARA prevalence of commercial farms was estimated at 6.6 ± 5.29% ranging from 0 to 38.3%. A subsequent linear mixed model was built to investigate the herd-level factors associated to higher within-herd SARA prevalence. Milking system, proportion of primiparous cows, herd size and seasons were all herd-level factors affecting SARA prevalence. Furthermore, milk production was positively, and milk fat yield negatively associated with SARA prevalence. Due to their moderate levels of accuracy, the SARA prediction models developed in our study, using data from continuous pH measurements on commercial farms, are not suitable for diagnostic purpose. However, these models can provide valuable information at the herd level.

Blood immune cell profiling in adults with longstanding type 1 diabetes is associated with macrovascular complications.

There is increasing evidence for heterogeneity in type 1 diabetes mellitus (T1D): not only the age of onset and disease progression rate differ, but also the risk of complications varies markedly. Consequently, the presence of different disease endotypes has been suggested. Impaired T and B cell responses have been established in newly diagnosed diabetes patients. We hypothesized that deciphering the immune cell profile in peripheral blood of adults with longstanding T1D may help to understand disease heterogeneity. Adult patients with longstanding T1D and healthy controls (HC) were recruited, and their blood immune cell profile was determined using multicolour flow cytometry followed by a machine-learning based elastic-net (EN) classification model. Hierarchical clustering was performed to identify patient-specific immune cell profiles. Results were compared to those obtained in matched healthy control subjects. Hierarchical clustering analysis of flow cytometry data revealed three immune cell composition-based distinct subgroups of individuals: HCs, T1D-group-A and T1D-group-B. In general, T1D patients, as compared to healthy controls, showed a more active immune profile as demonstrated by a higher percentage and absolute number of neutrophils, monocytes, total B cells and activated CD4+CD25+ T cells, while the abundance of regulatory T cells (Treg) was reduced. Patients belonging to T1D-group-A, as compared to T1D-group-B, revealed a more proinflammatory phenotype characterized by a lower percentage of FOXP3+ Treg, higher proportions of CCR4 expressing CD4 and CD8 T cell subsets, monocyte subsets, a lower Treg/conventional Tcell (Tconv) ratio, an increased proinflammatory cytokine (TNFα, IFNγ) and a decreased anti-inflammatory (IL-10) producing potential. Clinically, patients in T1D-group-A had more frequent diabetes-related macrovascular complications. Machine-learning based classification of multiparameter flow cytometry data revealed two distinct immunological profiles in adults with longstanding type 1 diabetes; T1D-group-A and T1D-group-B. T1D-group-A is characterized by a stronger pro-inflammatory profile and is associated with a higher rate of diabetes-related (macro)vascular complications.

Air Pollution Metabolomic Signatures and Chronic Respiratory Diseases Risk: A Longitudinal Study

BackgroundThough evidence has documented the associations of ambient air pollution with chronic respiratory diseases (CRD) and lung function, the underlying metabolic mechanisms remain largely unclear. Research QuestionHow does the metabolomic signature for air pollution relate to CRD risk, respiratory symptoms, and lung function? Study Design and MethodsWe retrieved 171,132 participants free of chronic obstructive pulmonary disease (COPD) and asthma at baseline from the UK Biobank, who had data on air pollution and metabolomics. Exposures to air pollutants (particulate matter with diameter ≤2.5 μm [PM2.5], PM10, nitrogen oxide [NOX], and NO2) were assessed for 4 years before baseline considering residential address histories. We used 10-fold cross-validation elastic net regression to identify air pollution-associated metabolites. Multivariable Cox models were used to assess the associations between metabolomic signatures and CRD risk. Mediation and pathway analysis were conducted to explore the metabolic mechanism underlying the associations. ResultsDuring a median follow-up of 12.51 years, 8,951 and 5,980 incident COPD and asthma cases were recorded. In multivariable Cox regressions, air pollution was positively associated with CRD risk (for example, hazard ratios [HR] per interquartile range [IQR] increment in PM2.5: 1.09; 95% confidence interval [CI]: 1.06-1.13). We identified 103, 86, 85, and 90 metabolites in response to PM2.5, PM10, NOX, and NO2 exposure, respectively. The metabolomic signatures showed significant associations with CRD risk (HR per standard deviation (SD) increment in PM2.5-metabolomic signature: 1.11; 95% CI: 1.09-1.14). Mediation analysis showed that peripheral inflammatory and erythrocyte-related markers mediated the effects of metabolomic signatures on CRD risk. We identified 14 and 12 perturbed metabolic pathways (energy metabolism and amino acid metabolism pathways, etc.) for PM2.5 and NOX-metabolomic signatures. InterpretationOur study identifies metabolomic signatures for air pollution exposure. The metabolomic signatures showed significant associations with CRD risk, and inflammatory and erythrocyte-related markers partly mediated the metabolomic signatures-CRD links.

The impact of digitalization on the economic growth of the European Union: an empirical study

The object of this study was the process of digitalization of the economy and society of the European Union. The task of researching the impact of digitalization on the economic growth of the European Union has been solved. The relevance of the chosen topic is due to the complexity of digital transformation processes taking place in the world economy and the economy of the European Union, in particular, the need to assess their directions and intensity. Four approaches to modeling the impact of digitalization on economic growth based on the neoclassical production function were proposed. The structure of the model was chosen, which makes it possible to evaluate the influence of various factors of digitalization on the key parameters of economic growth. A regression multivariate model for assessing the impact of digitalization on the economic growth of the European Union has been developed. To this end, a sequence of stages was performed: formation of the information base, grouping of digitalization factors, as well as their selection based on correlation analysis. With the help of regression analysis, the parameters of the production function were expressed through digitalization indicators. At the same time, due to the limitation of input data, the Elastic Net Regression method was used. This made it possible to ensure the quality of the new model, namely, to remove low-impact parameters, reduce the multicollinearity of factors, make the model statistically significant, and ensure the stability of coefficients to data changes. The resulting model is eleven-factor; it demonstrates a high predictive ability (the coefficient of determination is 0.987). It can be used as an analytical tool for assessing the impact of digitalization on the economic growth of the European Union. Practical use of the model will help governments and businesses make informed decisions about digital transformation policies

Circulating Metabolite Biomarkers of Glycemic Control in Youth-Onset Type 2 Diabetes.

We aimed to identify metabolites associated with loss of glycemic control in youth-onset type 2 diabetes. We measured 480 metabolites in fasting plasma samples from the TODAY (Treatment Options for Type 2 Diabetes in Adolescents and Youth) study. Participants (N = 393; age 10-17 years) were randomly assigned to metformin, metformin plus rosiglitazone, or metformin plus lifestyle intervention. Additional metabolomic measurements after 36 months were obtained in 304 participants. Cox models were used to assess baseline metabolites, interaction of metabolites and treatment group, and change in metabolites (0-36 months), with loss of glycemic control adjusted for age, sex, race, treatment group, and BMI. Metabolite prediction models of glycemic failure were generated using elastic net regression and compared with clinical risk factors. Loss of glycemic control (HbA1c ≥8% or insulin therapy) occurred in 179 of 393 participants (mean 12.4 months). Baseline levels of 33 metabolites were associated with loss of glycemic control (q < 0.05). Associations of hexose and xanthurenic acid with treatment failure differed by treatment randomization; youths with higher baseline levels of these two compounds had a lower risk of treatment failure with metformin alone. For three metabolites, changes from 0 to 36 months were associated with loss of glycemic control (q < 0.05). Changes in d-gluconic acid and 1,5-AG/1-deoxyglucose, but not baseline levels of measured metabolites, predicted treatment failure better than changes in HbA1c or measures of β-cell function. Metabolomics provides insight into circulating small molecules associated with loss of glycemic control and may highlight metabolic pathways contributing to treatment failure in youth-onset diabetes.

A Novel Hybrid Regression Model for Banking Loss Estimation

Given the critical need to identify financial risks in the banking sector early, this study presents a novel approach that uses historical financial ratios from the FDIC database to predict bank failures in the United States. Accurate estimation of potential losses is essential for risk management and decision-making procedures. We present a novel hybrid approach to loss estimation in the context of bank failures in this study. ElasticNet regression and relevant data extraction techniques are combined in our method to improve prediction accuracy. We conducted thorough experiments and evaluated our hybrid approach's performance against that of conventional regression techniques. With a remarkably low Mean Squared Error (MSE) of 0.001, a significantly high R-squared value of 0.98, and an Explained Variance Score of 0.95, our proposed model demonstrates superior performance compared to existing methodologies. The accuracy of our method is further demonstrated by the Mean Absolute Error (MAE) of 1200 units. Our results highlight the potential of our hybrid approach to transform loss estimation in the banking and finance domain, offering superior predictive capabilities and more accurate loss estimations.

High-throughput seed quality analysis in faba bean: leveraging Near-InfraRed spectroscopy (NIRS) data and statistical methods

Near-infrared spectroscopy (NIRS) provides a high-throughput phenotyping technique to assist breeding for improved faba bean seed quality. We combined chemical analysis of protein, oil content (and composition) with NIRS through chemometrics, employing Partial Least Squares (PLS), Elastic Net (EN), Memory-based Learning (MBL), and Bayes B (BB) as prediction models. Protein was the most reliably predicted trait (R2 = 0.96–0.98) across field trials, followed by oil (R2 = 0.82–0.86) and oleic acid (R2 = 0.31–0.68). Samples for training the models were selected using K-means clustering. The optimal statistical approach for prediction was compound-specific: PLS for protein (Root Mean Squared Error - RMSE = 0.46), BB for oil (RMSE = 0.067), and EN for oleic acid content (RMSE = 2.83). Reduced training set simulations revealed different effects on prediction accuracy depending on the model and compound. Several NIR regions were pinpointed as highly informative for the compounds, using the shrinkage and variable selection capabilities of EN and BB.

An elastic net regression model for predicting the risk of ICU admission and death for hospitalized patients with COVID-19

This study aimed to develop and validate prediction models to estimate the risk of death and intensive care unit admission in COVID-19 inpatients. All RT-PCR-confirmed adult COVID-19 inpatients admitted to Fujian Provincial Hospital from October 2022 to April 2023 were considered. Elastic Net Regression was used to derive the risk prediction models. Potential risk factors were considered, which included demographic characteristics, clinical symptoms, comorbidities, laboratory results, treatment process, prognosis. A total of 1906 inpatients were included finally by inclusion/exclusion criteria and were divided into derivation and test cohorts in a ratio of 8:2, where 1526 (80%) samples were used to develop prediction models under a repeated cross-validation framework and the remaining 380 (20%) samples were used for performance evaluation. Overall performance, discrimination and calibration were evaluated in the validation set and test cohort and quantified by accuracy, scaled Brier score (SbrS), the area under the ROC curve (AUROC), and Spiegelhalter-Z statistics. The models performed well, with high levels of discrimination (AUROCICU [95%CI]: 0.858 [0.803,0.899]; AUROCdeath [95%CI]: 0.906 [0.850,0.948]); and good calibrations (Spiegelhalter-ZICU: − 0.821 (p-value: 0.412); Spiegelhalter-Zdeath: 0.173) in the test set. We developed and validated prediction models to help clinicians identify high risk patients for death and ICU admission after COVID-19 infection.

Plasma metals, genetic risk, and rapid kidney function decline among type 2 diabetes

BackgroundRapid kidney function decline (RKFD) is a main clinical feature of early chronic kidney disease (CKD) in type 2 diabetes (T2D). Environmental and genetic factors influencing RKFD remain inadequately elucidated. ObjectivesThis study aimed to examine the associations of metals with RKFD among T2D and to further investigate the effect of metal mixtures on RKFD with the modifying effect of genetic susceptibility. MethodsThis study included 2209 people with T2D (1942 had genotyping data) free of CKD at baseline from the Dongfeng-Tongji cohort. We used inductively coupled plasma-mass spectrometry (ICP-MS) to measure 23 metals in baseline plasma. Using elastic net (ENET), multivariate logistic regression, and Bayesian kernel machine regression (BKMR) model, we examined independent associations of multiple metals with RKFD. We calculated the environmental risk score (ERS) to assess the effects of metal mixtures on RKFD and the genetic risk score (GRS) to assess genetic susceptibility. RKFD was defined as estimated glomerular filtration rate (eGFR) loss > 3 mL/min/1.73 m2/year. ResultsDuring a median of 9.8 years follow-up, 262 participants developed RKFD. Aluminum, vanadium, zinc, selenium, rubidium, tin, barium, and tungsten were screened from ENET. In multivariate logistic models, vanadium, selenium, and tungsten were negatively associated with RKFD, while zinc, tin, and rubidium were positively associated. The BKMR showed a nonlinear association of vanadium and rubidium with RKFD and interactions between metals (barium‑vanadium, barium‑rubidium). The ERS was positive associated with RKFD (per SD increase in ERS, OR = 1.94, 95% CI: 1.66, 2.27). No significant interaction between ERS and GRS was observed on RKFD, however, participants in the highest ERS and GRS group had the highest RKFD risk. ConclusionVanadium and rubidium were associated with RKFD in T2D. Metal mixtures was associated with an increased risk of RKFD in T2D, particularly in those at high genetic risk.

Machine learning to predict lymph node metastasis in T1 esophageal squamous cell carcinoma：a multicenter study.

Existing models do poorly when it comes to quantifying the risk of Lymph node metastases (LNM). This study aimed to develop a machine learning model for LNM in patients with T1 esophageal squamous cell carcinoma (ESCC). The study is multicenter, and population based. Elastic net regression (ELR), random forest (RF), extreme gradient boosting (XGB), and a combined (ensemble) model of these was generated. The contribution to the model of each factor was calculated. The models all exhibited potent discriminating power. The Elastic net regression performed best with externally validated AUC of 0.803, whereas the NCCN guidelines identified patients with LNM with an AUC of 0.576 and logistic model with an AUC of 0. 670. The most important features were lymphatic and vascular invasion and depth of tumor invasion. Models created utilizing machine learning approaches had excellent performance estimating the likelihood of LNM in T1 ESCC.

A computational framework to in silico screen for drug-induced hepatocellular toxicity.

Drug-induced liver injury (DILI) is the most common trigger for acute liver failure and the leading cause of attrition in drug development. In this study, we developed an in silico framework to screen drug-induced hepatocellular toxicity (INSIGHT) by integrating the post-treatment transcriptomic data from both rodent models and primary human hepatocytes. We first built an early prediction model using logistic regression with elastic net regularization for 123 compounds and established the INSIGHT framework that can screen for drug-induced hepatotoxicity. The 235 signature genes identified by INSIGHT were involved in metabolism, bile acid synthesis, and stress response pathways. Applying the INSIGHT to an independent transcriptomic dataset treated by 185 compounds predicted that 27 compounds show a high DILI risk, including zoxazolamine and emetine. Further integration with cell image data revealed that predicted compounds with high DILI risk can induce abnormal morphological changes in the endoplasmic reticulum and mitochondrion. Clustering analysis of the treatment-induced transcriptomic changes delineated distinct DILI mechanisms induced by these compounds. Our study presents a computational framework for a mechanistic understanding of long-term liver injury and the prospective prediction of DILI risk.

Predicting State-Level Firearm Suicide Rates: A Machine Learning Approach Using Public Policy Data

IntroductionOver 40,000 people die by suicide annually in the U.S., and firearms are the most lethal suicide method. There is limited evidence on the effectiveness of many state-level policies on reducing firearm suicide. The objective of this study was to identify public policies that best predict state-level firearm suicide rates. MethodsData from the Centers for Disease Control and Prevention's WONDER system and the State Firearm Law Database, a longitudinal catalog of 134 firearm safety laws, were analyzed. The analysis included 1,450 observations from 50 states spanning 1991-2019. An ElasticNet regression technique was used to analyze the relationship between the policy variables and firearm suicide rates. Nested cross-validation was performed to tune the model hyperparameters. The study data were collected and analyzed in 2023 and 2024. ResultsThe optimized ElasticNet approach had a mean MSE of 2.07, which was superior to non-regularized and dummy regressor models. The most influential policies for predicting the firearm suicide rate on average included laws requiring firearm dealers that sell handguns to have a state license and laws requiring individuals to obtain a permit to purchase a firearm through an approval process that includes law enforcement, among others. ConclusionsOn average, firearm suicide rates were lower in state-years that had each influential policy active. Notably, these analyses were ecological and non-causal. However, this study was able to use a supervised machine learning approach with inherent feature selection and many policy types to make predictions using unseen data (i.e., balancing Lasso and Ridge regularization penalties).

Metabolite Profiles of Plant-Based Diets and Cardiometabolic Risk in the Mediators of Atherosclerosis in South Asians Living in America Study

BackgroundHealthy plant-based diets have been associated with lower risk of type 2 diabetes (T2D). Metabolomics can be leveraged to identify potential pathways through which diet influences disease risk. ObjectivesThis study aimed to identify profiles of serum metabolites reflective of plant-based diets of varying quality and examine associations with cardiometabolic risk and T2D. MethodsWe included data from 687 participants of the Mediators of Atherosclerosis in South Asians Living in America (MASALA) cohort. An overall plant-based diet index (PDI), healthy PDI (hPDI), and unhealthy PDI (uPDI) were estimated from food frequency questionnaires. Serum metabolites were assayed using ultraperformance liquid chromatography mass spectrometry. Elastic net regression was used to identify sets of metabolites predictive of each diet index, and metabolite profile scores were calculated as the weighted sum of the selected metabolites. Cross-sectional associations between metabolite profile scores and cardiometabolic measures and prospective associations with incident T2D were evaluated with multivariable-adjusted linear and logistic regressions. ResultsMetabolite profiles for PDI, hPDI, and uPDI consisted of n = 51, 55, and 45 metabolites, respectively. Metabolites strongly positively correlated with diet indices included phosphatidylcholine (16:0/18:3) for PDI, phosphatidylethanolamine (20:1/20:4) and pantothenate for hPDI, and lysophosphatidylglycerol (18:2/0:0), proline, and lauric acid for uPDI. Higher metabolite profile scores for PDI and hPDI were associated with lower glycemia and lipids measures, whereas a higher uPDI metabolite score was associated with higher triglycerides and lower low density lipoprotein cholesterol and high density lipoprotein cholesterol. A higher metabolite score for hPDI was additionally associated with lower adiposity measures, higher liver fat attenuation, higher adiponectin, lower odds of overweight (odds ratio [OR]: 0.64; 95% confidence interval [CI]: 0.51, 0.81) and obesity (OR: 0.59; 95% CI: 0.48, 0.74), and lower odds of incident T2D (OR: 0.66; 95% CI: 0.45, 0.97). ConclusionsMetabolite profiles of different plant-based diets were identified. Metabolite profiles of overall and healthy plant-based diets were associated with favorable cardiometabolic risk profiles.

The prognostic importance of traumatic axonal injury on early MRI: the Trondheim TAI-MRI grading and quantitative models.

We analysed magnetic resonance imaging (MRI) findings after traumatic brain injury (TBI) aiming to improve the grading of traumatic axonal injury (TAI) to better reflect the outcome. Four-hundred sixty-three patients (8-70 years) with mild (n = 158), moderate (n = 129), or severe (n = 176) TBI and early MRI were prospectively included. TAI presence, numbers, and volumes at predefined locations were registered on fluid-attenuated inversion recovery (FLAIR) and diffusion-weighted imaging, and presence and numbers on T2*GRE/SWI. Presence and volumes of contusions were registered on FLAIR. We assessed the outcome with the Glasgow Outcome Scale Extended. Multivariable logistic and elastic-net regression analyses were performed. The presence of TAI differed between mild (6%), moderate (70%), and severe TBI (95%). In severe TBI, bilateral TAI in mesencephalon or thalami and bilateral TAI in pons predicted worse outcomes and were defined as the worst grades (4 and 5, respectively) in the Trondheim TAI-MRI grading. The Trondheim TAI-MRI grading performed better than the standard TAI grading in severe TBI (pseudo-R2 0.19 vs. 0.16). In moderate-severe TBI, quantitative models including both FLAIR volume of TAI and contusions performed best (pseudo-R2 0.19-0.21). In patients with mild TBI or Glasgow Coma Scale (GCS) score 13, models with the volume of contusions performed best (pseudo-R2 0.25-0.26). We propose the Trondheim TAI-MRI grading (grades 1-5) with bilateral TAI in mesencephalon or thalami, and bilateral TAI in pons as the worst grades. The predictive value was highest for the quantitative models including FLAIR volume of TAI and contusions (GCS score <13) or FLAIR volume of contusions (GCS score ≥ 13), which emphasise artificial intelligence as a potentially important future tool. The Trondheim TAI-MRI grading reflects patient outcomes better in severe TBI than today's standard TAI grading and can be implemented after external validation. The prognostic importance of volumetric models is promising for future use of artificial intelligence technologies. Traumatic axonal injury (TAI) is an important injury type in all TBI severities. Studies demonstrating which MRI findings that can serve as future biomarkers are highly warranted. This study proposes the most optimal MRI models for predicting patient outcome at 6 months after TBI; one updated pragmatic model and a volumetric model. The Trondheim TAI-MRI grading, in severe TBI, reflects patient outcome better than today's standard grading of TAI and the prognostic importance of volumetric models in all severities of TBI is promising for future use of AI.

IMG-20. IDENTIFICATION OF PRE-OPERATIVE IMAGING FEATURES ASSOCIATED WITH COGNITIVE IMPAIRMENT IN CHILDREN WITH BRAIN TUMOURS USING A LANGUAGE MODELLING SYSTEM COMBINED WITH MACHINE LEARNING ANALYSIS

Abstract BACKGROUND Cognitive impairments caused by paediatric brain tumours negatively impact future quality-of-life. Impairments of processing speed and working memory are important contributors to challenges these children encounter in education. Inconsistent image acquisition and poor imaging quality can restrict advanced radiomic analysis in historical datasets. We developed and refined a language modelling human image-analysis support system, passing data to specific machine learning techniques that can utilise datasets with many data points in a limited population. METHODS Medical records of children treated for brain tumours with surgery and radiotherapy in Manchester between 2000-2020 were reviewed with imaging accessible from 2006-2020. Pre-operative MRI scans were reviewed for 52 patients along with their most recent processing speed and/or working memory test scores. 274 radiological variables analysing tumour features and damage to brain structures were reported per patient using a standardised reporting tool. Univariate analysis was performed using Linear Regression with time-to-test analysed as a covariate. Multivariate analysis used Elastic Net Regression. RESULTS Univariate analysis revealed risk factors for slower processing speed included midline shift, tumour in the corpus callosum, right temporal periventricular white matter oedema, diffusion changes in the left frontal lobe or amygdala and FLAIR changes in the left caudate nucleus (p&amp;lt;0.005). Risk factors for impaired working memory included midline shift, cortical tumour location, left periventricular white matter oedema and diffusion or FLAIR changes in the left frontal lobe (p&amp;lt;0.005). Despite the limited population size and review of only pre-operative features, multivariate analysis produced multiple predictive models able to outperform chance when predicting processing speed and working memory outcomes. CONCLUSIONS This study provides evidence that a radiological language modelling approach can be utilised for outcome prediction. It identified imaging features predictive of cognitive impairments prior to surgery and radiotherapy. Further work is needed to assess inter-observer variability and validate these findings.

Evaluating the Improvement of Blend Potency Measurements in the Feed Frame of a Rotary Tablet Press Using Combined NIR and Raman Spectroscopy.

This study investigated the added value of combining both near-infrared (NIR) and Raman spectroscopy into a single NIRaman Combi Fiber Probe for in-line blend potency determination in the feed frame of a rotary tablet press. A five-component platform formulation was used, containing acetylsalicylic acid as the Active Pharmaceutical Ingredient (API). Calibration models for the determination of 1 and 5%w/w label claim tablets were developed using NIR and Raman spectra of powder blends ranging from 0.75 to 1.25%w/w and 3.75 to 6.25%w/w API, respectively. Step-change experiments with deliberate 10% deviation steps from the label claims were performed, from which the collected spectra were used for model validation. For model development and validation, low-level data fusion was explored through concatenation of preprocessed NIR and Raman spectra. Mid-level data fusion was also evaluated, based on extracted features of the preprocessed data. Herewith, score vectors were extracted by transforming preprocessed spectra through Principal Component Analysis, followed by critical feature selection through Elastic Net Regression. Partial Least Squares regression was applied to regress singular, low-level or mid-level fused data versus blend potency. It could be concluded that irrespective of the data fusion technique, an increase in Step-Change Sensitivity (SCS) and decrease in Root Mean Squared Error (RMSE) was observed when predicting the 5%w/w step-change experiment. For the prediction of the 1%w/w step-change experiment, no added benefit with regard to SCS and RMSE was observed due to the addition of the noisy NIR spectra.

IMG-10. IMAGING SUBTYPES OF PEDIATRIC LOW-GRADE GLIOMA ARE ASSOCIATED WITH DISTINCT MOLECULAR CHARACTERISTICS

Abstract BACKGROUND Pediatric low-grade gliomas (pLGGs) have diverse molecular subtypes with distinct prognoses, necessitating personalized treatments. We aimed to determine whether radiomic features of pLGGs co-segregate with molecular subtypes, establishing a proof-of-concept that distinct molecular aberrations could manifest in differentiable imaging features. METHODS Leveraging radiomic, genomic, and transcriptomic data from 201 pLGG patients from the Children’s Brain Tumor Network (CBTN), we identified imaging-based pLGG subtypes through their radiophenotypes. This involved investigating the association of imaging subtypes with 2021 WHO pLGG classifications and profiling transcriptional characteristics distinguishing pLGG clusters. We used principal component analysis and K-Means clustering on 881 radiomic features and incorporated clinical variables (age, sex, tumor location), and derived three imaging subtypes. RESULTS BRAF V600E mutations were mainly found in subtype 3, while BRAF::KIAA1549 fusion in subtype 1. To ascertain transcriptome pathways most closely associated with imaging-based subtypes, we applied a supervised machine learning model with elastic net logistic regression using a one-versus-rest strategy for each imaging cluster, considering the top 100 most variable pathways, identified through gene set enrichment and gene co-expression network analyses, molecular subtype, tumor location, age, sex, and race. The predictive accuracy (area under the receiver operating characteristic curve) of the model was 0.86 | 0.72 for imaging subtype 1, 0.93 | 0.71 for subtype 2, and 0.91 | 0.64 for subtype 3 in the training | testing sets. Notably, each imaging subtype exhibited distinct molecular mechanisms; subtype 1 showed up-regulation of oxidative phosphorylation, PDGFRB, and interleukin 4, 10, and 13 signaling, while subtype 3 was linked to histone acetylation and DNA methylation pathways, relating to BRAF V600E pLGGs. CONCLUSIONS Our radiogenomics study suggests that intrinsic molecular characteristics of tumors drive distinct imaging subgroups of pLGG, providing a basis for future multi-modal investigations that may better define progression and targetability of the disease.

MDB-42. RADIOMIC TEXTURAL AND MORPHOLOGICAL-BASED FEATURES TO RISK-STRATIFY PEDIATRIC MEDULLOBLASTOMA PATIENTS

Abstract BACKGROUND One key determinate in the treatment pathway for pediatric medulloblastoma (MB), the most frequent malignant brain tumor in children, is accurate risk-stratification. MB tumors are classified as standard- or high-risk based on current approaches (Chang’s/molecular stratification). However, there is still a need for additional attributes towards reliable risk-stratification. This work presents a radiomic-based approach that combines textural- and morphological-based attributes to risk-stratify MB patients. Our rationale is that coupling texture and morphology attributes from the intra-tumoral regions that quantify the heterogeneity and disorderly nature of aggressive tumors can transcend the current clinical approaches in accurately stratifying tumors into high- and standard-risk. METHODS T1-weighted MRI scans of 119 MB patients (2–18 years) were collected from Cincinnati Children’s Hospital Medical Center (Site1-n=42), Children’s Hospital Los Angeles (Site2-n=47), and Children’s Hospital of Philadelphia (Site3-n=30), used interchangeably for training and testing. Following preprocessing and performing annotations for the enhancing lesion and edema sub-compartments, 232 textural and morphological features were extracted from each sub-compartment. Namely, 214 textural (gradient, Haralick, intensity, Gabor, Laws, entropy) as well as 34 morphological (4 surface-based (e.g., curvedness and sharpness), 18 global (e.g., roundness, elongation, compactness, flatness)) features were extracted. Features were then fed into Elastic-Net regression models for survival analysis. RESULTS Our analysis revealed that texture and morphology features, combined, yielded Concordance index (CI) and p-value of 0.52, 0.0001, respectively, for the enhancing lesion, using Sites1,3 for training, and CI=0.61, p=0.0065 using Site3 for testing. Additionally, for edema, CI and p-value of 0.514, 0.0001 were obtained when employing Sites2,3 for training and CI=0.61, p=0.00096 using Site1 for testing. Results were not significant when using Chang’s or molecular stratification alone for survival analysis. CONCLUSIONS Our study shows that radiomic-based morphological and textural features show promise towards reliable MB risk-stratification.

Prediction of post-delivery hemoglobin levels with machine learning algorithms

Predicting postpartum hemorrhage (PPH) before delivery is crucial for enhancing patient outcomes, enabling timely transfer and implementation of prophylactic therapies. We attempted to utilize machine learning (ML) using basic pre-labor clinical data and laboratory measurements to predict postpartum Hemoglobin (Hb) in non-complicated singleton pregnancies. The local databases of two academic care centers on patient delivery were incorporated into the current study. Patients with preexisting coagulopathy, traumatic cases, and allogenic blood transfusion were excluded from all analyses. The association of pre-delivery variables with 24-h post-delivery hemoglobin level was evaluated using feature selection with Elastic Net regression and Random Forest algorithms. A suite of ML algorithms was employed to predict post-delivery Hb levels. Out of 2051 pregnant women, 1974 were included in the final analysis. After data pre-processing and redundant variable removal, the top predictors selected via feature selection for predicting post-delivery Hb were parity (B: 0.09 [0.05–0.12]), gestational age, pre-delivery hemoglobin (B:0.83 [0.80–0.85]) and fibrinogen levels (B:0.01 [0.01–0.01]), and pre-labor platelet count (B*1000: 0.77 [0.30–1.23]). Among the trained algorithms, artificial neural network provided the most accurate model (Root mean squared error: 0.62), which was subsequently deployed as a web-based calculator: https://predictivecalculators.shinyapps.io/ANN-HB. The current study shows that ML models could be utilized as accurate predictors of indirect measures of PPH and can be readily incorporated into healthcare systems. Further studies with heterogenous population-based samples may further improve the generalizability of these models.